Unanimous AI study says small AI-linked teams beat Polymarket in NBA forecasting

A new study presented June 9 at HAXD26 in Valencia found that 25-to-30 person forecasting teams using Unanimous AI’s platform reached 62% accuracy on 50 NBA against-the-spread picks, topping Polymarket’s 55%. The researchers say the result points to a faster, more explainable alternative to prediction markets for enterprise decisions and group forecasting.

Why it matters: - The study suggests small human teams linked by AI agents can outperform much larger prediction markets while also explaining their reasoning. - Unanimous AI says the approach could help organizations make faster decisions on risk, priorities, forecasts and alignment inside real-time meetings. - The findings add to a broader push to use AI to amplify human judgment instead of replacing it.

What happened: - Unanimous AI and Carnegie Mellon presented a paper at the 2nd International Conference on Human-AI Interaction and Experience Design, or HAXD26, in Valencia, Spain, on June 9, 2026. - The paper, “Conversational Forecasting Across Large Human Groups Using a Swarm of Surrogate AI Agents,” tested teams of 25 to 30 randomly selected sports fans on 50 NBA games against the spread. - Those teams used Unanimous AI’s Thinkscape collaboration platform, powered by Hyperchat AI technology, for five minutes per game. - The teams reached 62% accuracy, above the 50% baseline implied by Vegas odds and above Polymarket’s 55% accuracy on the same set of games. - The research paper was published by IEEE, with a version available here: the paper.

The details: - During each discussion, AI agents processed human input in real time and combined arguments, reasoning and sentiment into a probabilistic forecast. - The system also produced the reasoning behind each forecast, including the most influential arguments and the strongest counterarguments. - The study says sustained accuracy above 55% in against-the-spread sports forecasting is considered world-class handicapping. - Unanimous AI says the 62% result beat that benchmark by seven percentage points. - The company says prediction markets like Polymarket collect input in sequence, which can create momentum that overshoots or undershoots before converging. - The company says Conversational Swarms work in parallel, which helps the strongest deliberative arguments rise faster. - The study also argues that prediction markets require too many participants, take too long to converge and provide little insight into why the group reached a conclusion. - The conference theme was “Beyond Prompts: Designing Human-AI Teamwork.” - The event ran June 9-12, 2026, and was technically co-sponsored by the IEEE Spain Section. - A video of Dr. Louis Rosenberg’s talk is available here: Rosenberg’s talk.

Between the lines: - The results are not just about sports betting accuracy; they are a proof point for a workflow where AI organizes human judgment rather than substitutes for it. - If the findings hold up across other domains, the model could matter more for business and government decision-making than for prediction markets alone. - The comparison also reframes scale: the study says a few dozen coordinated people can outperform hundreds or thousands of independent traders when the AI layer structures the discussion well. - Unanimous AI is positioning this approach as “hyper-communication,” a term the company uses for large-scale real-time human conversations optimized by AI agents. - The company says its technologies are used by large organizations worldwide, including Fortune 1000 companies and the U.S. Air Force.

What’s next: - Unanimous AI is pointing to a growing body of peer-reviewed research on networked human teams connected by Hyperchat AI agents. - The company is continuing to market Thinkscape and related tools for organizations that need faster group decisions and richer output from meetings. - More publications are available here: research and publications. - More information about Thinkscape is available here: Thinkscape. - More information about Unanimous AI is available here: Unanimous AI.

The bottom line: - The study argues that small, AI-linked human teams can beat prediction markets on both accuracy and explanation, which could reshape how organizations think about collective intelligence.